Shulou(Shulou.com)11/24 Report--

Nvidia made a big move again, this time directly exploding the market with Super GPU GH200.

At yesterday's COMPUTEX conference, Nvidia CEO Huang Renxun announced to the world--

We have reached the tipping point of generative AI. From then on, there will be computing needs in every corner of the world.

Nvidia, whose share price has just soared by $200 billion, is ready for this moment.

At the beginning, Lao Huang, dressed in black leather, stepped onto the stage impassively and said, "Hello, everyone!" We're back! "

Then it came out with a "super GPU" GH200, and announced that Google Cloud, Meta and Microsoft would be the first to get GH200.

It is said that more than 3500 people came to the scene to experience the two-hour passionate speech.

Four years later, Lao Huang, who has been away for a long time, is also skyrocketing in Chinese.

GH200, the "super chip", wants to say that the highlight of this speech is still on GPU. After all, AI's iPhone has arrived.

Lao Huang held a chip in his left and right hand and announced that the "GH200 super chip" had been put into full production.

This "super GPU" uses NVLink-c2c interconnect technology to combine ARM-based energy-saving GraceCPU with high-performance NVIDIA H100 Tensor Core GPU to provide a total bandwidth of up to 900GB/s.

At present, more than 400 system configurations have been added to the system supported by GH200.

These system configurations are powered by different combinations of Nvidia's latest CPU, GPU and DPU architectures.

These include Grace, Hopper, Ada Lovelace, and BlueField, which are architectures created to meet the growing demand for generative AI.

In addition, Lao Huang also announced an even bigger one: the arrival of a 256-GH200 overcount.

Super DGX GH200, which went public this year, Nvidia said that the new DGX GH200 artificial intelligence supercomputing platform is designed for large-scale generation of AI load.

The supercomputer, made up of 256 Grace Hopper super chips, will have up to 1 exaflop of extraordinary AI performance and 144TB shared memory (nearly 500x more than the previous generation DGX A100).

For example, in GPT-3 training, it can be 2.2 times faster than the previous generation of DGX H100 clusters.

In addition, the behemoth contains 150 miles of optical fiber and more than 2000 fans.

So far, Nvidia has partnered with three giants, Google, Meta and Microsoft.

Due to the explosive growth of generative artificial intelligence, giants such as Microsoft and Google want to have more powerful and better performance systems.

The DGX H200 is designed to provide maximum throughput for large-scale scalability of maximum workloads by using Nvidia's custom NVLink Switch chips, bypassing the limitations of standard cluster connections such as InfiniBand and Ethernet.

In addition, Nvidia says it is building its own large AI supercomputer, NVIDIA Helios, which is expected to be launched this year.

It will use four DGX GH200 systems connected to the NVIDIA Quantum-2 InfiniBand network to improve data throughput and train large AI models.

In the past, the data center is very large and based on CPU, the iteration of the algorithm takes a long time, and most of the algorithms are CPU-centric.

Now, with Grace Hopper, it only takes a few days or even hours to complete the process. It's going to revolutionize the whole industry!

(wait a minute, isn't the parameter of PaLM 540B? )

Lao Huang: the more you buy, the more money you save! As the current carrier, such a 65-pound, $200000 H100 computer is the first computer in the world to be equipped with Transformer Engine, and it is also the most expensive computer in the world.

'it can be said that the more you buy products like this, the more you will save, 'Mr. Huang said.

Next, Lao Huang mentioned IBM 360 in 1964, emphasizing the importance of CPU.

Lao Huang repeated confidently, "60 years later, we now have a data center." Today, the data center is a computer. "

As Lao Huang said, a new computing model is being created.

Why is using GPU better than using CPU?

Lao Huang gave an analysis in terms of configuration: at a cost of US $1000, you can build a data center with 960 CPU, but this data center needs the power of 11GWh and handles the amount of data of 1x LLM (large language model).

But for the same amount of money, you can build a data center with 48 GPU, but as long as the power consumption of 3.2GWh, and can handle 44x LLM data volume.

You know, this configuration is amazing enough. However, this is not enough.

In order to achieve the ultimate performance, you can directly increase the number of GPU to 172 at the same power consumption.

At this time, the computing power can be as high as 150 times that of the CPU data center. Of course, the budget has also been raised to $34 million.

In addition, if you just want to finish the task at hand (1x LLM), Lao Huang will also help you cut down the cost--

For as little as $400,000, you can buy a data center with two GPU and consume only 0.13GWh.

There was a round of applause from the audience, and Lao Huang took out the mantra "The more you buy,The more you save" and even repeated it three times.

What is the strategy behind this? Lao Huang gave a formula.

MGX: modular architecture at the same time, Lao Huang also launched NVIDIA MGXTM, a reference architecture for system manufacturers to build more than 100 server variants quickly and cheaply.

It is said that this specification can reduce development costs by up to 3/4 and reduce development time by 2/3, which takes only six months.

With MGX, technology companies can optimize the basic system architecture for accelerated computing for their servers, and then choose their own GPU,DPU and CPU.

MGX can also be easily integrated into the cloud and enterprise data centers.

In addition to the hardware, MGX is supported by Nvidia's complete software stack, which enables developers and enterprises to build and accelerate AI, HPC and other applications.

This includes the software layer of the NVIDIA AI Enterprise,NVIDIA AI platform, which is characterized by more than 100 frameworks, pre-trained models and development tools to accelerate artificial intelligence and data science and provide full support for enterprise artificial intelligence development and deployment.

The highlight of the presentation was the introduction of AI into the game, the NPC role of real-time voice chat, and the new custom AI model contract manufacturing service-Avatar Cloud Engine (ACE) for Game.

At the scene, Lao Huang holds a RTX 4060 Ti in his right hand and a computer in his left hand, showing Cyberpunk 2077 running real-time ray tracing.

In a "cyberpunk" style ramen shop scene, the player presses a button to speak in his own voice, and the shopkeeper Jin will answer.

Jin is a NPC character, but his answer is generated in real time by the generative AI based on the player's voice input. Jin also has realistic facial animation and sound, consistent with the player's tone and backstory.

This realistic persona is generated using a real-time artificial intelligence model rendering tool Nvidia Ace.

Lao Huang said that the characters in the game are not pre-set. They have a typical task provider NPC type.

But from the video, you can see that the virtual character's conversation is a little stiff, but not too bad.

Those without AI expertise will be abandoned. Over the past 40 years, we have created PC, the Internet, mobile, the cloud, and this is the age of artificial intelligence.

What would you create? Whatever it is, chase it like we do. Run, don't go. Either you run for food, or you allow yourself to escape and become food.

On May 27, Huang Renxun delivered a graduation speech at National Taiwan University.

At the moment, he is attracting the attention of the world.

Instantly become the leader of trillion yuan, so that his words are more confident.

Huang Renxun said that every company and individual should be familiar with artificial intelligence, otherwise, there is a risk of failure.

He stressed: agile companies will use artificial intelligence to improve their status, such companies will not fail.

Many people worry that AI will take away their jobs, but it is the people who have mastered AI technology that will really take your jobs.

At that time, he predicted in his speech: in all respects, the prosperity of AI is a rebirth opportunity for the computer industry. In the next decade, our industry will use new AI computers to replace trillions of dollars worth of traditional computers.

And from today's speech, we seem to have glimpsed the embryonic form of this future.

Reference:

Https://www.youtube.com/watch?v=fHwmLOYJU_w

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.